arch/powerpc/platforms/pseries/ras.c

   1 /*
   2  * Copyright (C) 2001 Dave Engebretsen IBM Corporation
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  17  */
  18
  19 /* Change Activity:
  20  * 2001/09/21 : engebret : Created with minimal EPOW and HW exception support.
  21  * End Change Activity
  22  */
  23
  24 #include <linux/errno.h>
  25 #include <linux/threads.h>
  26 #include <linux/kernel_stat.h>
  27 #include <linux/signal.h>
  28 #include <linux/sched.h>
  29 #include <linux/ioport.h>
  30 #include <linux/interrupt.h>
  31 #include <linux/timex.h>
  32 #include <linux/init.h>
  33 #include <linux/slab.h>
  34 #include <linux/pci.h>
  35 #include <linux/delay.h>
  36 #include <linux/irq.h>
  37 #include <linux/random.h>
  38 #include <linux/sysrq.h>
  39 #include <linux/bitops.h>
  40
  41 #include <asm/uaccess.h>
  42 #include <asm/system.h>
  43 #include <asm/io.h>
  44 #include <asm/pgtable.h>
  45 #include <asm/irq.h>
  46 #include <asm/cache.h>
  47 #include <asm/prom.h>
  48 #include <asm/ptrace.h>
  49 #include <asm/machdep.h>
  50 #include <asm/rtas.h>
  51 #include <asm/udbg.h>
  52 #include <asm/firmware.h>
  53
  54 #include "ras.h"
  55
  56 static unsigned char ras_log_buf[RTAS_ERROR_LOG_MAX];
  57 static DEFINE_SPINLOCK(ras_log_buf_lock);
  58
  59 char mce_data_buf[RTAS_ERROR_LOG_MAX];
  60
  61 static int ras_get_sensor_state_token;
  62 static int ras_check_exception_token;
  63
  64 #define EPOW_SENSOR_TOKEN       9
  65 #define EPOW_SENSOR_INDEX       0
  66 #define RAS_VECTOR_OFFSET       0x500
  67
  68 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id,
  69                                         struct pt_regs * regs);
  70 static irqreturn_t ras_error_interrupt(int irq, void *dev_id,
  71                                         struct pt_regs * regs);
  72
  73 /* #define DEBUG */
  74
  75
  76 static void request_ras_irqs(struct device_node *np,
  77                         irqreturn_t (*handler)(int, void *, struct pt_regs *),
  78                         const char *name)
  79 {
  80         int i, index, count = 0;
  81         struct of_irq oirq;
  82         const u32 *opicprop;
  83         unsigned int opicplen;
  84         unsigned int virqs[16];
  85
  86         /* Check for obsolete "open-pic-interrupt" property. If present, then
  87          * map those interrupts using the default interrupt host and default
  88          * trigger
  89          */
  90         opicprop = get_property(np, "open-pic-interrupt", &opicplen);
  91         if (opicprop) {
  92                 opicplen /= sizeof(u32);
  93                 for (i = 0; i < opicplen; i++) {
  94                         if (count > 15)
  95                                 break;
  96                         virqs[count] = irq_create_mapping(NULL, *(opicprop++));
  97                         if (virqs[count] == NO_IRQ)
  98                                 printk(KERN_ERR "Unable to allocate interrupt "
  99                                        "number for %s\n", np->full_name);
 100                         else
 101                                 count++;
 102
 103                 }
 104         }
 105         /* Else use normal interrupt tree parsing */
 106         else {
 107                 /* First try to do a proper OF tree parsing */
 108                 for (index = 0; of_irq_map_one(np, index, &oirq) == 0;
 109                      index++) {
 110                         if (count > 15)
 111                                 break;
 112                         virqs[count] = irq_create_of_mapping(oirq.controller,
 113                                                             oirq.specifier,
 114                                                             oirq.size);
 115                         if (virqs[count] == NO_IRQ)
 116                                 printk(KERN_ERR "Unable to allocate interrupt "
 117                                        "number for %s\n", np->full_name);
 118                         else
 119                                 count++;
 120                 }
 121         }
 122
 123         /* Now request them */
 124         for (i = 0; i < count; i++) {
 125                 if (request_irq(virqs[i], handler, 0, name, NULL)) {
 126                         printk(KERN_ERR "Unable to request interrupt %d for "
 127                                "%s\n", virqs[i], np->full_name);
 128                         return;
 129                 }
 130         }
 131 }
 132
 133 /*
 134  * Initialize handlers for the set of interrupts caused by hardware errors
 135  * and power system events.
 136  */
 137 static int __init init_ras_IRQ(void)
 138 {
 139         struct device_node *np;
 140
 141         ras_get_sensor_state_token = rtas_token("get-sensor-state");
 142         ras_check_exception_token = rtas_token("check-exception");
 143
 144         /* Internal Errors */
 145         np = of_find_node_by_path("/event-sources/internal-errors");
 146         if (np != NULL) {
 147                 request_ras_irqs(np, ras_error_interrupt, "RAS_ERROR");
 148                 of_node_put(np);
 149         }
 150
 151         /* EPOW Events */
 152         np = of_find_node_by_path("/event-sources/epow-events");
 153         if (np != NULL) {
 154                 request_ras_irqs(np, ras_epow_interrupt, "RAS_EPOW");
 155                 of_node_put(np);
 156         }
 157
 158         return 0;
 159 }
 160 __initcall(init_ras_IRQ);
 161
 162 /*
 163  * Handle power subsystem events (EPOW).
 164  *
 165  * Presently we just log the event has occurred.  This should be fixed
 166  * to examine the type of power failure and take appropriate action where
 167  * the time horizon permits something useful to be done.
 168  */
 169 static irqreturn_t
 170 ras_epow_interrupt(int irq, void *dev_id, struct pt_regs * regs)
 171 {
 172         int status = 0xdeadbeef;
 173         int state = 0;
 174         int critical;
 175
 176         status = rtas_call(ras_get_sensor_state_token, 2, 2, &state,
 177                            EPOW_SENSOR_TOKEN, EPOW_SENSOR_INDEX);
 178
 179         if (state > 3)
 180                 critical = 1;  /* Time Critical */
 181         else
 182                 critical = 0;
 183
 184         spin_lock(&ras_log_buf_lock);
 185
 186         status = rtas_call(ras_check_exception_token, 6, 1, NULL,
 187                            RAS_VECTOR_OFFSET,
 188                            irq_map[irq].hwirq,
 189                            RTAS_EPOW_WARNING | RTAS_POWERMGM_EVENTS,
 190                            critical, __pa(&ras_log_buf),
 191                                 rtas_get_error_log_max());
 192
 193         udbg_printf("EPOW <0x%lx 0x%x 0x%x>\n",
 194                     *((unsigned long *)&ras_log_buf), status, state);
 195         printk(KERN_WARNING "EPOW <0x%lx 0x%x 0x%x>\n",
 196                *((unsigned long *)&ras_log_buf), status, state);
 197
 198         /* format and print the extended information */
 199         log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0);
 200
 201         spin_unlock(&ras_log_buf_lock);
 202         return IRQ_HANDLED;
 203 }
 204
 205 /*
 206  * Handle hardware error interrupts.
 207  *
 208  * RTAS check-exception is called to collect data on the exception.  If
 209  * the error is deemed recoverable, we log a warning and return.
 210  * For nonrecoverable errors, an error is logged and we stop all processing
 211  * as quickly as possible in order to prevent propagation of the failure.
 212  */
 213 static irqreturn_t
 214 ras_error_interrupt(int irq, void *dev_id, struct pt_regs * regs)
 215 {
 216         struct rtas_error_log *rtas_elog;
 217         int status = 0xdeadbeef;
 218         int fatal;
 219
 220         spin_lock(&ras_log_buf_lock);
 221
 222         status = rtas_call(ras_check_exception_token, 6, 1, NULL,
 223                            RAS_VECTOR_OFFSET,
 224                            irq_map[irq].hwirq,
 225                            RTAS_INTERNAL_ERROR, 1 /*Time Critical */,
 226                            __pa(&ras_log_buf),
 227                                 rtas_get_error_log_max());
 228
 229         rtas_elog = (struct rtas_error_log *)ras_log_buf;
 230
 231         if ((status == 0) && (rtas_elog->severity >= RTAS_SEVERITY_ERROR_SYNC))
 232                 fatal = 1;
 233         else
 234                 fatal = 0;
 235
 236         /* format and print the extended information */
 237         log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, fatal);
 238
 239         if (fatal) {
 240                 udbg_printf("Fatal HW Error <0x%lx 0x%x>\n",
 241                             *((unsigned long *)&ras_log_buf), status);
 242                 printk(KERN_EMERG "Error: Fatal hardware error <0x%lx 0x%x>\n",
 243                        *((unsigned long *)&ras_log_buf), status);
 244
 245 #ifndef DEBUG
 246                 /* Don't actually power off when debugging so we can test
 247                  * without actually failing while injecting errors.
 248                  * Error data will not be logged to syslog.
 249                  */
 250                 ppc_md.power_off();
 251 #endif
 252         } else {
 253                 udbg_printf("Recoverable HW Error <0x%lx 0x%x>\n",
 254                             *((unsigned long *)&ras_log_buf), status);
 255                 printk(KERN_WARNING
 256                        "Warning: Recoverable hardware error <0x%lx 0x%x>\n",
 257                        *((unsigned long *)&ras_log_buf), status);
 258         }
 259
 260         spin_unlock(&ras_log_buf_lock);
 261         return IRQ_HANDLED;
 262 }
 263
 264 /* Get the error information for errors coming through the
 265  * FWNMI vectors.  The pt_regs' r3 will be updated to reflect
 266  * the actual r3 if possible, and a ptr to the error log entry
 267  * will be returned if found.
 268  *
 269  * The mce_data_buf does not have any locks or protection around it,
 270  * if a second machine check comes in, or a system reset is done
 271  * before we have logged the error, then we will get corruption in the
 272  * error log.  This is preferable over holding off on calling
 273  * ibm,nmi-interlock which would result in us checkstopping if a
 274  * second machine check did come in.
 275  */
 276 static struct rtas_error_log *fwnmi_get_errinfo(struct pt_regs *regs)
 277 {
 278         unsigned long errdata = regs->gpr[3];
 279         struct rtas_error_log *errhdr = NULL;
 280         unsigned long *savep;
 281
 282         if ((errdata >= 0x7000 && errdata < 0x7fff0) ||
 283             (errdata >= rtas.base && errdata < rtas.base + rtas.size - 16)) {
 284                 savep = __va(errdata);
 285                 regs->gpr[3] = savep[0];        /* restore original r3 */
 286                 memset(mce_data_buf, 0, RTAS_ERROR_LOG_MAX);
 287                 memcpy(mce_data_buf, (char *)(savep + 1), RTAS_ERROR_LOG_MAX);
 288                 errhdr = (struct rtas_error_log *)mce_data_buf;
 289         } else {
 290                 printk("FWNMI: corrupt r3\n");
 291         }
 292         return errhdr;
 293 }
 294
 295 /* Call this when done with the data returned by FWNMI_get_errinfo.
 296  * It will release the saved data area for other CPUs in the
 297  * partition to receive FWNMI errors.
 298  */
 299 static void fwnmi_release_errinfo(void)
 300 {
 301         int ret = rtas_call(rtas_token("ibm,nmi-interlock"), 0, 1, NULL);
 302         if (ret != 0)
 303                 printk("FWNMI: nmi-interlock failed: %d\n", ret);
 304 }
 305
 306 int pSeries_system_reset_exception(struct pt_regs *regs)
 307 {
 308         if (fwnmi_active) {
 309                 struct rtas_error_log *errhdr = fwnmi_get_errinfo(regs);
 310                 if (errhdr) {
 311                         /* XXX Should look at FWNMI information */
 312                 }
 313                 fwnmi_release_errinfo();
 314         }
 315         return 0; /* need to perform reset */
 316 }
 317
 318 /*
 319  * See if we can recover from a machine check exception.
 320  * This is only called on power4 (or above) and only via
 321  * the Firmware Non-Maskable Interrupts (fwnmi) handler
 322  * which provides the error analysis for us.
 323  *
 324  * Return 1 if corrected (or delivered a signal).
 325  * Return 0 if there is nothing we can do.
 326  */
 327 static int recover_mce(struct pt_regs *regs, struct rtas_error_log * err)
 328 {
 329         int nonfatal = 0;
 330
 331         if (err->disposition == RTAS_DISP_FULLY_RECOVERED) {
 332                 /* Platform corrected itself */
 333                 nonfatal = 1;
 334         } else if ((regs->msr & MSR_RI) &&
 335                    user_mode(regs) &&
 336                    err->severity == RTAS_SEVERITY_ERROR_SYNC &&
 337                    err->disposition == RTAS_DISP_NOT_RECOVERED &&
 338                    err->target == RTAS_TARGET_MEMORY &&
 339                    err->type == RTAS_TYPE_ECC_UNCORR &&
 340                    !(current->pid == 0 || is_init(current))) {
 341                 /* Kill off a user process with an ECC error */
 342                 printk(KERN_ERR "MCE: uncorrectable ecc error for pid %d\n",
 343                        current->pid);
 344                 /* XXX something better for ECC error? */
 345                 _exception(SIGBUS, regs, BUS_ADRERR, regs->nip);
 346                 nonfatal = 1;
 347         }
 348
 349         log_error((char *)err, ERR_TYPE_RTAS_LOG, !nonfatal);
 350
 351         return nonfatal;
 352 }
 353
 354 /*
 355  * Handle a machine check.
 356  *
 357  * Note that on Power 4 and beyond Firmware Non-Maskable Interrupts (fwnmi)
 358  * should be present.  If so the handler which called us tells us if the
 359  * error was recovered (never true if RI=0).
 360  *
 361  * On hardware prior to Power 4 these exceptions were asynchronous which
 362  * means we can't tell exactly where it occurred and so we can't recover.
 363  */
 364 int pSeries_machine_check_exception(struct pt_regs *regs)
 365 {
 366         struct rtas_error_log *errp;
 367
 368         if (fwnmi_active) {
 369                 errp = fwnmi_get_errinfo(regs);
 370                 fwnmi_release_errinfo();
 371                 if (errp && recover_mce(regs, errp))
 372                         return 1;
 373         }
 374
 375         return 0;
 376 }